Knowledge graph application method based on big data

Abstract

The invention discloses a knowledge graph application method based on big data. The method comprises the following steps: step 1, identifying a target entity and a search intention which a user hopes to search from a query submitted by the user; 2, matching the trigger word of the set target entity with the abstract trigger word of the event class, extracting personalized features based on the matched target entity, and predicting the preference of the user; 3, performing multi-angle recommendation of each vertical field based on the knowledge graph architecture; step 4, establishing an application rule of a multi-vertical field, applying the target entity obtained from the query and the knowledge graph of the matched event item to the multi-vertical field, generating an interest prediction value recommended by the user to the event item according to a user feature vector and an event item feature vector extracted from the knowledge graph by the user, and calculating the interest prediction value of the user to the event item. The method has the characteristics of accurate matching and strong relevance recommendation.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph application technology, and specifically to a knowledge graph application method based on big data. Background Art

[0002] Most current search and recommendation systems are statistical models. Traditional search and recommendation models mostly generate results based on the user's historical behavior. The statistical laws of user behavior are formed from the user's historical search, click, shopping, and browsing behaviors. For users who do not obey the statistical laws, traditional statistical models will simply regard them as abnormalities. However, any user's behavior often has its motivation. Due to the lack of data, traditional models are difficult to capture the motivation behind individual behavior. They can only simply transplant the behavioral characteristics of the group to the individual. Therefore, the recommendation services provided by the existing recommendation systems are all the same, which easily obliterates the individual characteristics of users or objects. Although there are also a large number of personalized models, they only refine the scope of application of statistical models to a certain extent, and it is difficult to change their essence of using group behavioral characteristics to serve individuals. Therefore, traditional statistical models mostly generate results based on historical behavior. For users who do not conform to the statistical laws, they will be simply regarded as abnormalities, and the motivation behind the user cannot be obtained. It is difficult to understand the user's intention and it is difficult to accurately match. Therefore, it is necessary to design a knowledge graph application system based on big data for accurate matching and strong correlation recommendation. Summary of the invention

[0003] The purpose of the present invention is to provide a knowledge graph application method based on big data to solve the problems raised in the above background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a knowledge graph application method based on big data, comprising the following steps: Step 1: Identify the target entity and search intent that the user wants to find from the query submitted by the user; Step 2: Match the trigger words of the set target entity with the abstract trigger words of the event class, extract personalized features based on the matched target entity, and predict the user's preference; Step 3: Each vertical field makes multi-angle recommendations based on the knowledge graph architecture; Step 4: Establish application rules for multiple vertical fields and apply the knowledge graph of target entities obtained from the query and matched event items to multiple vertical fields.

[0005] According to the above technical solution, the step of identifying the target entity and search intent that the user wants to find from the query submitted by the user includes: Obtain the user's query data on the system's search page, analyze the web page and jump to the user's main search interface using the user's query keywords, customize concept entities and the relationships between entities according to the application field, use data attributes to describe the entity's own attribute information, and find the target entity and its related content in the existing knowledge graph through the set query statement; Obtain the corresponding information intent and transaction intent based on the target entity of the user's access query and its related content.

[0006] According to the above technical solution, the step of matching the trigger word of the set target entity with the abstract trigger word of the event class includes: Set the trigger word of the output target entity and match the trigger word with the abstract trigger word of the event class; When the matching results match, continue to match the object elements and environmental elements of the target entity; When the matching result does not match, continue to search and match the next event class in the event knowledge graph; When the trigger words, object elements, and environmental elements of an event all match a certain event class in the event knowledge graph, the event class is stored and the next event class in the event knowledge graph is matched; When there are multiple event classes in the final matched event class set and there is a classification relationship between the event classes, find the lowest-level sub-event class and use it as the event class to which the event instance belongs; When the final matching event class set is empty, it is judged that the target entity cannot find a corresponding event class in the event knowledge graph.

[0007] According to the above technical solution, the step of extracting personalized features based on the matched target entity and predicting the user's preference includes: Preprocess the target entity data and search intent data output by the matching; Based on the interaction between items and entities, an interaction feature matrix is ​​formed, and each interaction matrix is ​​combined with the associated entity to obtain the interaction matrix and knowledge graph based on associated semantics corresponding to the target entity; The associated data information of user nodes is integrated on the knowledge graph to obtain the user's personalized features; Aggregate the associated data information of the interactive nodes in the knowledge graph to obtain the personalized features of the user's associated interactive data; Perform alternate feature extraction on user personalized features and user associated interaction data features to obtain user feature vectors and user associated interaction feature vectors respectively; Predict user preferences through prediction functions.

[0008] According to the above technical solution, the steps of performing multi-angle recommendations in each vertical field based on the knowledge graph architecture include: The final feature vector representation of the user target entity and event item is extracted from the knowledge graph, and the preference is predicted by performing inner product operation on the user feature vector and the event item feature vector. Each value in the multidimensional vector can be regarded as one type of evaluation value, that is, the first value in the vector represents an event item corresponding to the user entity target. The larger the value, the higher the user's preference for the related event item. The larger the value of the event item feature vector, the higher the recommendation value of the event item.

[0009] According to the above technical solution, the step of applying the knowledge graph of the target entity obtained from the query and the matched event-type items to multiple vertical fields includes: Construct knowledge graph application rules for multiple vertical fields, based on the acquisition of target entity data in different fields, all based on the data information generated by users' operations on the web query end of the corresponding vertical field. After serializing the collected data set in this field, the data is further cleaned and preprocessed to construct knowledge triples that conform to the knowledge links in the knowledge graph of this field, and then converted into target json data after feature extraction. The json data is transmitted to the front and back ends for logical interaction, and visualization for knowledge graph applications is formed based on interaction and matching predictions.

[0010] According to the above technical solution, the knowledge graph application system includes: An entity semantic association module is used to identify the semantic association between the target entity that the user wants to find and the corresponding event item from the query submitted by the user; The knowledge graph analysis and matching module is used to analyze and match the trigger words of the set target entity with the abstract trigger words of the event class using the knowledge graph; The knowledge graph application module is used to apply the knowledge graph of the target entities obtained from the query and the matched event-type items to multiple vertical fields.

[0011] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention identifies the target entity and search intent that the user wants to find from the query submitted by the user, matches the trigger word of the set target entity with the abstract trigger word of the event class, and extracts personalized features based on the matched target entity, predicts the user's preference, and enables each vertical field to make multi-angle recommendations in the field based on the knowledge graph architecture; when establishing application rules for multiple vertical fields, the knowledge graph of the target entity obtained from the query and the matched event class project is applied to multiple vertical fields, so that the user generates a user's interest prediction value for the event project recommendation based on the user feature vector extracted from the knowledge graph and the event project feature vector, so that the recommendation match for the target entity is more accurate and the recommendation relevance is stronger. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 A flowchart of a knowledge graph application method based on big data provided in Example 1 of the present invention; Figure 2 A schematic diagram of the module composition of a knowledge graph application system based on big data provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] Embodiment 1: Figure 1 This is a flowchart of a knowledge graph application method based on big data provided in Example 1 of the present invention. This embodiment can be applied to the scenario of knowledge graph application. This method can be executed by a knowledge graph application system based on big data provided in this embodiment, such as Figure 1 As shown, the method specifically comprises the following steps: Step 1: Identify the target entity and search intent that the user wants to find from the query submitted by the user; In an embodiment of the present invention, the user's query data is obtained on the search page of the system, and the user's search main interface is jumped to by analyzing the web page and using the user's query keywords as the medium. Concept entities and the relationships between entities are customized according to the application field, and the attribute information of the entity itself is described using data attributes. The target entity and its related content are found in the existing knowledge graph through the set query statement. By giving a query represented by a set of entities, the entity most relevant to the query is found, and a concept is generated to explain the potential connection between the related entity and the entity in the query. The target is optimized by finding the most relevant entity, so that the introduction of the entity most relevant to the query can maintain the semantics of the original entity in the query as much as possible, the found results are sorted by entity, and then reasonably classified according to the result content, and presented to the user in an organized manner, so as to expand the related content beyond the target entity, and effectively present the related entities to the user to improve the search diversity.

[0015] Exemplarily, the corresponding information intent and transaction intent are obtained based on the target entity of the user's access query and its related content, that is, the information about a certain topic that the user wants to obtain is constructed, specifically the transaction activity information using the network as the medium, including shopping and downloading Internet resources, and then the subject classification of the entity search is solved by sorting the candidate entity categories, and grammatical analysis is performed based on the classified subjects, and the part-of-speech tagging or grammatical modification relationship of the keywords in the query is output, and the semantic roles corresponding to the keywords in the query are output, so as to obtain the semantic associations between the query keywords.

[0016] Step 2: Match the trigger words of the set target entity with the abstract trigger words of the event class, extract personalized features based on the matched target entity, and predict the user's preference; In an embodiment of the present invention, a trigger word of the output target entity is set, and the trigger word is matched with the abstract trigger word of the event class. Only after the matching results match, the object elements and environmental elements of the target entity can be matched. If they do not match, the target entity is considered to be inconsistent with the event class, and the next event class is searched and matched in the event knowledge graph. The event object element matching and event environmental element matching are the same as above. When the trigger word, object element, and environmental element of the event can all match an event class in the event knowledge graph, the event class is stored, and the next event class in the event knowledge graph is matched. If there are multiple event classes in the final matched event class set and there is a classification relationship between the event classes, find the bottom-level sub-event class and use it as the event class to which the event instance belongs. If the final matched event class set is empty, it is judged that the target entity cannot find a corresponding event class in the event knowledge graph.

[0017] Exemplarily, the target entity data and search intention data of the matching output are preprocessed. Since the recommended predicted items can correspond to the entities of the knowledge graph, different items can describe the same target entity. Therefore, an interaction feature matrix is ​​formed based on the interaction between the items and the entities, and each interaction matrix is ​​combined with the associated entity to obtain an interaction matrix and a knowledge graph based on associated semantics corresponding to the target entity. The interaction matrix includes users and their historical click data, and the knowledge graph includes user nodes corresponding to the historical click data. The associated data information of the user nodes is fused on the knowledge graph to obtain the personalized features of the users. The associated data information of the interaction nodes is aggregated in the knowledge graph to obtain the personalized features of the user-associated interaction data. Alternate feature extraction is performed on the user personalized features and the user-associated interaction data features to obtain user feature vectors and user-associated interaction feature vectors, respectively. Multiple iterative operations are performed to make the waveform of the user personalized features interact with the waveform of the user-associated interaction personalized features, so as to more deeply explore the user feature vectors and the user-associated interaction feature vectors, perform inner product operations on the user feature vectors and the user-associated interaction feature vectors, and predict the user's preferences through prediction functions.

[0018] Step 3: Each vertical field makes multi-angle recommendations based on the knowledge graph architecture; In an embodiment of the present invention, the final feature vector representation of the user target entity and the event item is extracted from the knowledge graph, and the preference is predicted by performing an inner product operation on the user feature vector and the event item feature vector. Each value in the multidimensional vector can be regarded as one type of evaluation value, that is, the first value in the vector represents a certain event item corresponding to the user entity target. The larger the value, the higher the user's preference for the related event item, and the larger the value of the event item feature vector, the higher the recommendation value of the event item. Therefore, after calculating the inner product, the larger the values ​​of the user feature vector and the event item feature vector, the larger the inner product result, which means that the user prefers the target entity based on the knowledge graph according to the search. The greater the interest level of the recommended corresponding event item, the smaller the interest level is. The inner product operation is performed on the user representation vector and the event item representation vector, and the predicted interest value of the event item corresponding to the user target entity is obtained through the inner product operation. The predicted value at this time is the value within the entire recommendation range. By limiting the range of the output value, the output value obtained by the inner product operation is limited to between 0 and 1 using an activation function. The predicted interest value obtained at this time is the user's click probability for the event item. The greater the click probability, the more likely the user is to interact with the event item. Therefore, based on the user feature vector and event item feature vector extracted from the knowledge graph, a user's interest prediction value for the event item recommendation is generated.

[0019] Exemplarily, the knowledge graph architecture model outputs the interaction value of the user's multi-target entities and multi-event items by inputting the path set of the queried user target entity and the matched event project, and forms a recommendation plan based on the interaction value. For the user's target entity and event project, the path between them is extracted from the knowledge graph, and the target entity and entity type in the path are projected into two embedding vectors. Since there are different relationships between entities, even if the head and tail entities are the same, the interaction relationship may be different. Therefore, the relationship between the two entities before and after the connection is also projected into an independent embedding vector. On the basis of the path represented by the embedding vector sequence, the neural network is used to memorize the long-term dependency in the order of the user's search input to serialize the data information. By calculating the embedded vector, the entire path representation of the target entity in the knowledge graph is obtained. After obtaining the path representation, the interaction score between the user target entity and the event item is calculated based on the path. For each path in the path set between the user target entity and the event item, its embedded vector is calculated separately to obtain a score set of the interaction between the user target entity and the event item. The interaction scores are aggregated into a total score. The higher the score, the greater the weight of the path. By setting the weight hyperparameter, while providing the user with the relevant event item recommendation results based on the target entity of the search query, the recommendation reason is provided based on the path information between the user target entity and the event item with the highest interaction score. Different path types correspond to different recommendation reasons.

[0020] Step 4: Establish application rules for multiple vertical fields and apply the knowledge graph of target entities obtained from the query and matched event items to multiple vertical fields.

[0021] In the embodiment of the present invention, the knowledge graph application rules of multiple vertical fields are constructed, and the target entity data of different fields are collected and acquired based on the data information formed by the user's operation on the web query terminal of the corresponding vertical field. After the collected data set of this field is serialized, the data is further cleaned and preprocessed to construct the knowledge triples that conform to the knowledge link in the knowledge graph of this field, and then converted into target json data after feature extraction, and the json data is transmitted to the front and back ends for logical interaction, and visualization for the knowledge graph application is formed according to the interaction and matching prediction; Exemplarily, users use Web service intelligent search to implement single keyword search, multi-keyword search, and demand-based search functions. The single keyword search function means that when a Web service keyword is entered, the system returns the Web service containing the entered keyword; in the multi-keyword search function, users select the function, provider, type and other related attribute information of the Web service through multiple tabs to obtain qualified Web services; demand search is when the user enters his or her target demand in the search box, the front-end interface transmits the demand to the back-end, and the back-end parses the user demand and uses the Web service recommendation algorithm to obtain the Web service that meets the user demand and returns it to the front-end. After obtaining the Web service that meets the target demand through the search interface, click on the relevant Web service to enter the Web service knowledge graph visualization interface. In the knowledge graph visualization interface, users can explore the relationship between Web services, filter Web service types, and other functions. The interface intuitively displays the relationship between multiple entities in the form of a knowledge graph.

[0022] Embodiment 2: Embodiment 2 of the present invention provides a knowledge graph application system based on big data. Figure 2 A schematic diagram of the module composition of a knowledge graph application system based on big data provided in the second embodiment of the present invention, such as Figure 2 As shown, the system includes: An entity semantic association module is used to identify the semantic association between the target entity that the user wants to find and the corresponding event item from the query submitted by the user; The knowledge graph analysis and matching module is used to analyze and match the trigger words of the set target entity with the abstract trigger words of the event class using the knowledge graph; The knowledge graph application module is used to apply the knowledge graph of the target entities obtained from the query and the matched event-type items to multiple vertical fields.

[0023] In some embodiments of the present invention, the entity semantic association module includes: The target entity query module is used to find the target entity and its related content in the existing knowledge graph through the set query statement; The search intent acquisition module is used to acquire the corresponding search intent according to the target entity and its related content of the user's access query; The keyword association module is used to query the semantic association information data between keywords.

[0024] In some embodiments of the present invention, the knowledge graph analysis and matching module includes: The event class matching module is used to match the trigger words of the target entity with the abstract trigger words of the event class; A feature extraction module, used for alternately extracting user personalized features and user associated interactive data features; User preference prediction module, used to predict user preferences through prediction functions; The multi-angle recommendation module is used to make multi-angle recommendations in various vertical fields based on the knowledge graph architecture.

[0025] In some embodiments of the present invention, the knowledge graph application module includes: Application rule establishment module, used to establish application rules for multiple vertical fields; Multi-domain application module, used for knowledge graph recommendation based on application rules in multiple fields; The visualization display module is used to form a visualization display for knowledge graph applications based on interactions and matching predictions.

[0026] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0027] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A knowledge graph application method based on big data, characterized by: The method comprises the following steps: Step 1: Identify the target entity and search intent that the user wants to find from the query submitted by the user; Step 2: Match the trigger words of the set target entity with the abstract trigger words of the event class, extract personalized features based on the matched target entity, and predict the user's preference; Step 3: Each vertical field makes multi-angle recommendations based on the knowledge graph architecture; Step 4: Establish application rules for multiple vertical fields, and apply the knowledge graph of target entities and matched event items obtained from the query to multiple vertical fields; The step of matching the trigger word of the set target entity with the abstract trigger word of the event class includes: Set the trigger word of the output target entity and match the trigger word with the abstract trigger word of the event class; When the matching results match, continue to match the object elements and environmental elements of the target entity; When the matching result does not match, continue to search and match the next event class in the event knowledge graph; When the trigger words, object elements, and environmental elements of an event all match a certain event class in the event knowledge graph, the event class is stored and the next event class in the event knowledge graph is matched; When there are multiple event classes in the final matched event class set and there is a classification relationship between the event classes, find the lowest-level sub-event class and use it as the event class to which the event instance belongs; When the final matching event class set is empty, it is judged that the target entity cannot find a corresponding event class in the event knowledge graph; The steps of extracting personalized features based on the matched target entity and predicting the user's preference include: Preprocess the target entity data and search intent data output by the matching; Based on the interaction between items and entities, an interaction feature matrix is ​​formed, and each interaction matrix is ​​combined with the associated entity to obtain the interaction matrix and knowledge graph based on associated semantics corresponding to the target entity; The associated data information of user nodes is integrated on the knowledge graph to obtain the user's personalized features; Aggregate the associated data information of the interactive nodes in the knowledge graph to obtain the personalized features of the user's associated interactive data; Perform alternate feature extraction on user personalized features and user associated interaction data features to obtain user feature vectors and user associated interaction feature vectors respectively; Predict user preferences through prediction functions; The step of identifying the target entity and search intent that the user wants to find from the query submitted by the user includes: Obtain the user's query data on the system's search page, jump to the user's main search interface by analyzing the web page and using the user's query keywords as a medium, customize concept entities and the relationship between entities according to the application field, use data attributes to describe the attribute information of the entity itself, and find the target entity and its related content in the existing knowledge graph through the set query statement; Obtain the corresponding information intent and transaction intent based on the target entity and related content of the user's access query; According to the big data-based knowledge graph application method of claim 1, it is characterized in that: the step of performing multi-angle recommendations in each vertical field based on the knowledge graph architecture includes: The final feature vector representation of the user target entity and event item is extracted from the knowledge graph, and the preference is predicted by performing inner product operation on the user feature vector and the event item feature vector. Each value in the multidimensional vector can be regarded as one type of evaluation value, that is, the first value in the vector represents an event item corresponding to the user entity target. The larger the value, the higher the user's preference for the related event item. The larger the value of the event item feature vector, the higher the recommendation value of the event item.

2. The knowledge graph application method based on big data according to claim 1 is characterized in that: The step of applying the knowledge graph of the target entity obtained from the query and the matched event-type items to multiple vertical fields includes: Construct knowledge graph application rules for multiple vertical fields, based on the acquisition of target entity data in different fields, all based on the data information generated by users' operations on the web query end of the corresponding vertical field. After serializing the collected data set in this field, the data is further cleaned and preprocessed to construct knowledge triples that conform to the knowledge links in the knowledge graph of this field, and then converted into target json data after feature extraction. The json data is transmitted to the front and back ends for logical interaction, and visualization for knowledge graph applications is formed based on interaction and matching predictions.

3. A knowledge graph application system for executing a knowledge graph application method based on big data as claimed in claim 2, characterized in that: The system comprises: An entity semantic association module is used to identify the semantic association between the target entity that the user wants to find and the corresponding event item from the query submitted by the user; The knowledge graph analysis and matching module is used to analyze and match the trigger words of the set target entity with the abstract trigger words of the event class using the knowledge graph; The knowledge graph application module is used to apply the knowledge graph of the target entities obtained from the query and the matched event-type items to multiple vertical fields.

4. The knowledge graph application system according to claim 3, characterized in that: The entity semantic association module includes: The target entity query module is used to find the target entity and its related content in the existing knowledge graph through the set query statement; The search intent acquisition module is used to acquire the corresponding search intent according to the target entity and its related content of the user's access query; The keyword association module is used to query the semantic association information data between keywords.

5. The knowledge graph application system according to claim 4, characterized in that: The knowledge graph analysis and matching module includes: The event class matching module is used to match the trigger words of the target entity with the abstract trigger words of the event class; A feature extraction module, used for alternately extracting user personalized features and user associated interactive data features; User preference prediction module, used to predict user preferences through prediction functions; The multi-angle recommendation module is used to make multi-angle recommendations in various vertical fields based on the knowledge graph architecture.

Images